UGENet: Learning Discriminative Embeddings for Unconstrained Gaze Estimation Network via Self-Attention Mechanism in Human-Computer Interaction
Hai Liu, Song Yu, Tingting Liu, Jianping Ju, Jianying Tang · 2024
Gaze estimation plays an important role for indicating the human behavior and intention. It has developed rapidly with the field of computer vision, enabling it to be used in a variety of applications, such as virtual reality and human-computer interaction. However, gaze estimation in the field is still challenging due to different head postures, lighting conditions and eye appearances. In this work, we propose a novel gaze estimation model with framework based on ResNet-50, which improves its estimation accuracy. In addition, the same loss function for each predicted angle is used to improve the generalization performance of the model. Finally, the proposed model is validated on the two most unconstrained popular datasets, such as MPIIGAZE and GAZE360. The experimental results show the proposed approach performs better compared to other better methods.